The Research Of Dynamic Load Balancing Strategy With PVM

With the progress of science and technology, people brings forwardhigher requests to computing ability. The traditional High PerformanceComputer couldn't satisfy the requirement of every walk of life andevery hierarchy in price performance ratio, applicability and therequirement to programmer. Network parallel system appeared with therapid progress of network technology, the popularization of personalcomputer, mini and workstation. The computing ability of the NetworkParallel Computing Circumstance combined by multi-computers maybeexceeds one high performance computer.Network Parallel System can be used as one single computingcircumstance, it is made up of a set of interlinked computing units andcorrelative resources on the special or the universal network. Dependingon its low price, short developing period, small investment risk,reduced system resources, good system expansibility and single systemimage, more and more researchers join the research of Network ParallelSystem.Many academies and scientific research departments invest muchmanpower and material resources in the research works of network parallelcomputing since its appearance. As one of its key technologies, loadbalancing technology has got more researchers's attention, JilinUniversity and Tsinghua University had got many fruits. Load balancingstrategies include static and dynamic load balancing strategies, theall-pervading method used in static load balancing is dichotomy, theresearch still should be deepened for dynamic load balancing.This paper puts forward one kind of self-adaptive dynamic loadbalancing strategies started by receiver and the load information prediction method of this strategies is based on simulated annealingalgorithm. We will write parallel programs in VC depending on PVM. Thedetailed research contents are below:(1) Finding and analyzing the merits and faults of the existing loadbalancing strategies, discussing their application area;(2)The building of network parallel computing environment and knowingthe support of pvm to load balancing deeply;(3)Put forward one integrated load index to measure the load of everyPC and prediction the load information by applying Neural Networkstechnology;(4)Designing one DLBS supporting heterogeneous environment basedon PVM, adopting the combined method of self-adaptive and receiverstarted;(5)System performance test in two different test circumstances.